CSCI 4717/5717 Computer Architecture

1
CSCI 4717/5717 Computer Architecture

• Topic Storage Media
• Reading Stallings, Chapter 6

2
Types of External Memory

• Magnetic Disk
• RAID
• Removable
• Optical
• CD-ROM
• CD-Recordable (CD-R)
• CD-R/W
• DVD
• Magnetic Tape
• Magnetic Disk

3
Physical Disk

• Disk substrate coated with magnetizable material
  (iron oxiderust)
• Substrate used to be aluminium now glass
• Improved surface uniformity -- Increases
  reliability
• Reduction in surface defects -- Reduced
  read/write errors
• Lower fly heights
• Better stiffness
• Better shock/damage resistance

4
Read and Write Mechanisms
5
Read and Write Mechanisms (continued)

• Recording and retrieval via conductive coil(s)
  called a head(s)
• May be single read/write head or separate ones
• During read/write, head is stationary (actually
  moves radially to platters) and platter rotates
  beneath head

6
Hard Drive Write

• Current through coil produces magnetic field
• Pulses sent to head
• Magnetic pattern recorded on surface below

7
Hard Drive Read (traditional)

• Magnetic field moving relative to coil produces
  current Analogous to a generator or alternator
• Coil can be the same for read and write
• Used with
• Floppies
• Older harddrives

8
Hard Drive Read (contemporary)

• Separate read head, close to write head
• Partially shielded magneto resistive (MR) sensor
• Electrical resistance depends on direction of
  magnetic field Passing current through it
  results in different voltage levels for different
  resistances
• High frequency operation -- Higher storage
  density and speed

9
Data Organization and Formatting
10
Data Organization and Formatting (continued)

• Concentric rings or tracks
• Track is same width as head
• Thousands of tracks per platter surface
• Intertrack gaps Gaps between tracks protect
  data integrity
• Reduce intertrack gap
• increase capacity
• possibly increase errors due to misalignment of
  head or interference from other tracks
• Constant angular velocity Same number of bits
  per track (variable packing density)

11
Tracks divided into sectors

• Minimum block size is one sector although may
  have more than one sector per block
• Typically hundreds of sectors per track
• May be fixed or variable in length
• Contemporary systems are fixed-length with 512
  bytes being common
• Sectors also have gaps called intratrack or
  intersector gaps

12
Constant Angular Velocity (CAV)

• Imagine a matrix with the rows as tracks and the
  columns as sectors.
• Twist matrix into a disk and see how much more
  packed the center is than the outside.
• Creates pie shaped sectors and concentric tracks
• Regardless of head position, sectors pass beneath
  it at the same (constant) speed
• Capacity limited by density on inside track
• Outer tracks waste with lower data density

13
Multiple Zone Recording
14
Multiple Zone Recording (continued)

• Divide disk into zones typical number is 16
• Each zone has fixed bits/sectors per track
• More complex circuitry to adjust for different
  data rates as heads move farther out.

15
Identifying SectorsST506 Example (old)
16
Formatting

• Two kinds of formatting
• Low level allows hard drive to find sectors
• O/S level allows for file system
• Must be able to identify start of track and
  sector
• Format disk
• Additional information not available to user
• Marks tracks and sectors

17
Characteristics of Hard Drives

• Head Motion
• Disk Portability
• Sides
• Platters
• Head Mechanism

18
Head Motion

• Fixed head vs. heads on a movable arm
• Fixed head (old)
• One read write head per track
• Heads mounted on fixed ridged arm
• Movable head
• Heads move radially across tracks
• One read write head per side

19
Disk Portability

• Removable vs. fixed
• Removable disk
• Examples floppy, ZIP, Jazz
• Can be removed from drive and replaced with
  another disk
• Provides unlimited storage capacity
• Easy data transfer between systems
• Non-removable disk permanently mounted in the
  drive

20
Sides and Platters

• Single (old or cheap) vs. double (typical) sided
• Single or multiple platter
• One head per sideHeads are joined and aligned
• Aligned tracks on each platter form cylinders
• Data is striped by cylinder
• reduces head movement
• Increases speed (transfer rate)

21
Cylinders
22
Head mechanism

• There are a number of characteristics of the
  head that affect drive performance
• Head size
• Distance of head from platter

23
Head Mechanism Tradeoffs

• Smaller heads allow for higher densities, but
  force head to be closer to the disk
• The closer the head, the greater risk of
  "crashes
• Distance of head from magnetic media
• Contact (Floppy)
• Fixed gap
• Flying (Winchester)
• Head rests on platter at rest
• When platter spins, air pressure lifts head from
  platter

24
Data Encoding

• Data is not stored as two directions of magnetic
  polarization corresponding to two values, 1 and
  0.
• Reasons
• Hard drive heads detect the changes in magnetic
  direction, not the direction of the field
• Difficult to read large blocks of all ones or all
  zeros eventually controller would lose
  synchronization
• One method for storing data uses a clock to
  define the bit positions, and by watching how the
  magnetic field changes with respect to that clock
  indicates presence of one or zero

25
FM Encoding

• A magnetic field change at the beginning and
  middle of a bit time represents a logic one
• A a magnetic field change only at the beginning
  represents a logic zero
• Referred to as Frequency Modulation (FM)

26
MFM Encoding

• Just like FM except that changes at beginning of
  bit time are removed unless two 0s are next to
  each other
• Called Modified Frequency Modulation (MFM)

27
RLL Encoding

• Goals of encoding
• to ensure enough polarity changes to maintain bit
  synchronization
• to ensure enough bit sequences are defined so
  that any sequence of ones and zeros can be
  handled and
• to allow for the highest number of bits to be
  represented with the fewest number of polarity
  changes

28
RLL Encoding (continued)

• Run Length Limited (RLL) uses polarity changes
  to represent sequences of bits rather than
  individual 0s or 1s

29
RLL Encoding (continued)

• Note that the shortest period between polarity
  changes is one and a half bit periods.
• This produces a 50 increased data density over
  MFM encoding.

30
Latest Encoding Technology

• Improved encoding methods have been introduced
  since the development of RLL
• Use digital signal processing and other methods
  to realize better data densities.
• These methods include Partial Response, Maximum
  Likelihood (PRML) and Extended PRML (EPRML)
  encoding.

31
S.M.A.R.T.

• Self-Monitoring, Analysis Reporting Technology
  System (S.M.A.R.T.) is a method used to predict
  hard drive failures
• Controller monitors hard drive functional
  parameters
• For example, longer spin-up times may indicate
  that the bearings are going bad
• S.M.A.R.T. enabled drives can provide an alert to
  the computer's BIOS warning of a parameter that
  is functioning outside of its normal range
• Attribute values are stored in the hard drive as
  an integer in the range from 1 to 253. The lower
  the value, the worse the condition is.
• Depending on the parameter and the manufacturer,
  different failure thresholds are set for each of
  the parameters.

32
Sample S.M.A.R.T. Parameters

• Power On Hours This indicates the age of the
  drive.
• Spin Up Time A longer spin up time may indicate
  a problem with the assembly that spins the
  platters.
• Temperature Higher temperatures also might
  indicate a problem with the assembly that spins
  the platters.
• Head Flying Height A reduction in the flying
  height of a Winchester head may indicate it is
  about to crash into the platters.
• Doesnt cover all possible failures IC failure
  or a failure caused by a catastrophic event

33
Speed

• Queuing time waiting for I/O device to be
  useable
• Waiting for device if device is serving another
  request
• Waiting for channel if device shares a channel
  with other devices (multiplexing)
• Disk rotating at a constant speed (energy saver
  disk may stop)

34
Seek time

• Process of finding data on a disk
• Find correct track by moving head (moveable head)
• Selecting head (fixed head) takes no time
• Some details cannot be pinned down
• Ramping functions
• Distance between current track and desired track
• Shorter distances and lighter components have
  reduced seek time

35
Rotational Latency

• Waiting for data to rotate under head
• Floppies 3600 RPM
• Hard Drives up to 15,000 RMP
• Average rotational delay is 1/2 time for full
  rotation
• Total Access time Seek Latency

36
Transfer Time

• Transfer time time it takes to retrieve the
  data as it passes under the head
• T b/(rN)
• where
• T transfer time
• b number of bytes to transfer
• N number of bytes on a track (i.e., bytes per
  full revolution)
• r rotation speed in RPS (i.e., tracks per
  second)

37
Rotational Position Sensing (RPS)

• Allows other devices to use I/O channel while
  seek is in process.
• When seek is complete, device predicts when data
  will pass under heads
• At a fixed time before data is expected to come,
  tries to re-establish communications with
  requesting processor if fails to reconnect,
  must wait full disk turn before new attempt is
  made RPS miss

38
Random access

• File is arranged in contiguous sectors only one
  seek time per track
• File is scattered to different sectors or device
  is shared with multiple processes seek time
  increased to once per sector

39
Redundant Array of Independent Disks (RAID)

• Rate of improvement in secondary storage has not
  kept up with that of processors or main memory
• In many system, gains can be had through parallel
  systems
• In disk systems, multiple requests can be
  serviced concurrently if there are multiple disks
  and the data for parallel requests is stored on
  different disks

40
RAID (continued)

• Standardization of multi-disk arrays
• 7 levels (0 through 6)
• Not a hierarchy
• Common characteristics
• Set of physical disks viewed as single logical
  drive by O/S
• Data distributed across multiple physical drives
  of array
• Can use redundant capacity to store parity
  information to aid in error correction/detection
• Third characteristic is needed because multiple
  mechanisms mean that there are more possibilities
  for failure

41
Striping

• User's data and applications see one logical
  drive
• Data is divided into strips
• Could be physical blocks, sectors, or some other
  unit
• The strips are then mapped to the different
  physical drives

42
Striping (continued)
43
RAID 0

• May not be considered RAID officially as it
  doesn't support third characteristic from above
  common characteristics No redundancy
• Data striped across all disks
• Round Robin striping
• Performance characteristics Increases speed
  since multiple data requests are probably in
  sequence of strips and therefore can be done in
  parallel (High I/O request rate)

44
RAID 0 (continued)
45
RAID 1

• Mirrored Disks 2 copies of each stripe on
  separate disks
• Data is striped across disks just like RAID 0
• Read from either slight performance increase 1
  disk has shorter seek time
• Write to both slight performance drop one disk
  will have longer seek time
• Recovery is simple swap faulty disk
  re-mirror no down time
• Performance characteristics Same as for RAID 0
• Expensive since twice capacity is required
  likely to be limited to critical system software
  and data files

46
RAID 1 (continued)
47
RAID 2

• Disks are synchronized to the point where each
  head is in same position on each disk
• On a single read or write, all disks are accessed
  simultaneously
• Striped at the bit level
• Error correction calculated across corresponding
  bits on disks
• Multiple parity disks store Hamming code w/parity
  (SEC-DED) error correction in corresponding
  position

48
RAID 2 (continued)

• Error correction is redundant as Hamming and such
  are already used within stored data.
• Only effective when many errors occur
• Lots of redundancy
• Expensive
• Not commercially accepted
• Performance characteristics Only one I/O request
  at a time (non-parallel)

49
RAID 2 (continued)
50
RAID 3

• Similar to RAID 2
• Only one redundant disk, no matter how large the
  array
• Simple parity bit for each set of corresponding
  bits doesn't actually detect failed drive, but
  can replace it
• Data on failed drive can be reconstructed from
  surviving data and parity info

51
RAID 3 (continued)

• Example, assume RAID 3 with 5 drives
• X4(i) X3(i) ? X2(i) ? X1(i) ? X0(i)
• Failed bit (e.g., X1(i)) can be replaced with
• X1(i) X4(i) ? X3(i) ? X2(i) ? X0(i)
• Equation derived from XOR'ing X4(i) ? X1(i) to
  both sides.
• Performance characteristics Very high transfer
  rates
• Problem Only one I/O request at a time
  (non-parallel)

52
RAID 3 (continued)
53
RAID 4

• Not commercially accepted
• Each disk operates independently
• Large stripes
• Bit-by-bit parity calculated across stripes on
  each disk stored on parity disk
• Performance characteristics
• High I/O request rates (parallel)
• Less suited for high data transfer rates

54
RAID 4 (continued)

• Problem there is a write penalty with each
  write
• old data strip must be read
• old parity strip must be read
• a new parity strip must be calculated
• a new parity strip must be stored
• new data must be stored

55
RAID 4 (continued)

• Original parity calculation
• X4(i) X3(i) ? X2(i) ? X1(i) ? X0(i)
• New bit is stored (e.g., X1(i)) parity is
  recalculated
• X4'(i) X3(i) ? X2(i) ? X1'(i) ? X0(i)
• X4'(i) X3(i) ? X2(i) ? X1'(i) ? X0(i) ? X1(i) ?
  X1(i)
• X4'(i) X4(i) ? X1(i) ? X1'(i)

56
RAID 4 (continued)
57
RAID 5

• Like RAID 4 except drops parity disk
• Parity strips are staggered across all data disks
• Round robin allocation for parity stripe
• Avoids RAID 4 bottleneck at parity disk
• Commonly used in network servers

58
RAID 5 (continued)
59
RAID 6

• Two parity calculations
• XOR parity is one of them
• Independent data check algorithm
• Stored in separate blocks on different disksUser
  requirement of N disks needs N2
• High data availability
• Three disks need to fail for data loss
• Significant write penalty

60
RAID 6 (continued)
61
RAID Summary